Support Windows as remote node. (#2431)

test/config/examples/mnist-keras.yml

@@ -14,7 +14,6 @@ assessor:
 trial:
   codeDir: ../../../examples/trials/mnist-keras
   command: python3 mnist-keras.py --num_train 200 --epochs 1
-  gpuNum: 0
 
 useAnnotation: false
 multiPhase: false

test/config/examples/mnist-nested-search-space.yml

@@ -15,7 +15,6 @@ assessor:
 trial:
   codeDir: ../../../examples/trials/mnist-nested-search-space
   command: python3 mnist.py --batch_num 10
-  gpuNum: 0
 
 useAnnotation: false
 multiPhase: false

test/config/examples/mnist-pytorch.yml

@@ -14,7 +14,6 @@ assessor:
 trial:
   codeDir: ../../../examples/trials/mnist-pytorch
   command: python3 mnist.py  --epochs 1 --batch_num 10
-  gpuNum: 0
 
 useAnnotation: false
 multiPhase: false

test/config/examples/mnist-tfv1.yml

@@ -14,7 +14,6 @@ assessor:
 trial:
   codeDir: ../../../examples/trials/mnist-tfv1
   command: python3 mnist.py --batch_num 10
-  gpuNum: 0
 
 useAnnotation: false
 multiPhase: false

test/config/integration_tests.yml

 
 defaultTestCaseConfig:
-  launchCommand: nnictl create --config $configFile
+  launchCommand: nnictl create --config $configFile --debug
   stopCommand: nnictl stop
   experimentStatusCheck: True
   platform: linux darwin win32
@@ -22,7 +22,7 @@ testCases:
   validator:
 
   # launch command, default launch command is 'nnictl create --config $configFile'
-  launchCommand: nnictl create --config $configFile
+  launchCommand: nnictl create --config $configFile --debug
 
   # stop command, default stop command is 'nnictl stop', empty means no stop command
   stopCommand: nnictl stop
@@ -38,15 +38,24 @@ testCases:
 
 - name: mnist-tfv1
   configFile: test/config/examples/mnist-tfv1.yml
+  config:
+    maxTrialNum: 1
+    trialConcurrency: 1
 
 - name: mnist-keras
   configFile: test/config/examples/mnist-keras.yml
+  config:
+    maxTrialNum: 2
+    trialConcurrency: 1
 
 - name: mnist-pytorch
   configFile: test/config/examples/mnist-pytorch.yml
 
 - name: mnist-annotation
   configFile: test/config/examples/mnist-annotation.yml
+  config:
+    maxTrialNum: 1
+    trialConcurrency: 1
 
 - name: cifar10-pytorch
   configFile: test/config/examples/cifar10-pytorch.yml

test/nni_test/nnitest/naive_test.py

@@ -10,7 +10,7 @@ import sys
 import time
 import traceback
 
-from utils import is_experiment_done, get_experiment_id, get_nni_log_path, read_last_line, remove_files, setup_experiment, detect_port, snooze
+from utils import is_experiment_done, get_experiment_id, get_nni_log_path, read_last_line, remove_files, setup_experiment, detect_port, wait_for_port_available
 from utils import GREEN, RED, CLEAR, EXPERIMENT_URL
 
 NNI_SOURCE_DIR = '..'
@@ -71,7 +71,7 @@ def naive_test(args):
     assert assessor_result == expected, 'Bad assessor result'
 
     subprocess.run(['nnictl', 'stop'])
-    snooze()
+    wait_for_port_available(8080, 10)
 
 def stop_experiment_test(args):
     config_file = args.config
@@ -86,19 +86,20 @@ def stop_experiment_test(args):
     experiment_id = get_experiment_id(EXPERIMENT_URL)
     proc = subprocess.run(['nnictl', 'stop', experiment_id])
     assert proc.returncode == 0, '`nnictl stop %s` failed with code %d' % (experiment_id, proc.returncode)
-    snooze()
+    wait_for_port_available(8080, 10)
     assert not detect_port(8080), '`nnictl stop %s` failed to stop experiments' % experiment_id
 
     # test cmd `nnictl stop --port`
     proc = subprocess.run(['nnictl', 'stop', '--port', '8990'])
     assert proc.returncode == 0, '`nnictl stop %s` failed with code %d' % (experiment_id, proc.returncode)
-    snooze()
+    wait_for_port_available(8990, 10)
     assert not detect_port(8990), '`nnictl stop %s` failed to stop experiments' % experiment_id
 
     # test cmd `nnictl stop --all`
     proc = subprocess.run(['nnictl', 'stop', '--all'])
     assert proc.returncode == 0, '`nnictl stop --all` failed with code %d' % proc.returncode
-    snooze()
+    wait_for_port_available(8888, 10)
+    wait_for_port_available(8989, 10)
     assert not detect_port(8888) and not detect_port(8989), '`nnictl stop --all` failed to stop experiments'
 
 

test/nni_test/nnitest/run_tests.py

@@ -15,7 +15,7 @@ import ruamel.yaml as yaml
 from utils import get_experiment_status, get_yml_content, dump_yml_content, get_experiment_id, \
     parse_max_duration_time, get_trial_stats, deep_update, print_trial_job_log, get_failed_trial_jobs, \
     get_experiment_dir, print_experiment_log
-from utils import GREEN, RED, CLEAR, STATUS_URL, TRIAL_JOBS_URL, EXPERIMENT_URL, REST_ENDPOINT, detect_port
+from utils import GREEN, RED, CLEAR, STATUS_URL, TRIAL_JOBS_URL, EXPERIMENT_URL, REST_ENDPOINT, wait_for_port_available
 import validators
 
 it_variables = {}
@@ -157,7 +157,7 @@ def launch_test(config_file, training_service, test_case_config):
             if num_failed > 0:
                 print('failed jobs: ', num_failed)
                 break
-            time.sleep(3)
+            time.sleep(1)
     except:
         print_experiment_log(experiment_id=experiment_id)
         raise
@@ -189,16 +189,6 @@ def case_included(name, cases):
             return True
     return False
 
-def wait_for_port_available(port, timeout):
-    begin_time = time.time()
-    while True:
-        if not detect_port(port):
-            return
-        if time.time() - begin_time > timeout:
-            msg = 'port {} is not available in {} seconds.'.format(port, timeout)
-            raise RuntimeError(msg)
-        time.sleep(5)
-
 def match_platform(test_case_config):
     return sys.platform in test_case_config['platform'].split(' ')
 

test/nni_test/nnitest/utils.py

@@ -168,6 +168,13 @@ def detect_port(port):
     except:
         return False
 
-def snooze():
-    '''Sleep to make sure previous stopped exp has enough time to exit'''
-    time.sleep(6)
+
+def wait_for_port_available(port, timeout):
+    begin_time = time.time()
+    while True:
+        if not detect_port(port):
+            return
+        if time.time() - begin_time > timeout:
+            msg = 'port {} is not available in {} seconds.'.format(port, timeout)
+            raise RuntimeError(msg)
+        time.sleep(1)

test/pipelines/pipelines-it-remote-linux-to-windows.yml

+jobs:
+  - job: "integration_test_remote_linux_to_windows"
+    timeoutInMinutes: 120
+    steps:
+      - script: make clean
+        displayName: "clean nni source code"
+      - task: CopyFilesOverSSH@0
+        inputs:
+          sshEndpoint: $(end_point)
+          contents: |
+            **
+            !**/dist/**
+            !**/node_modules/**
+          targetFolder: /tmp/nnitest/$(Build.BuildId)
+          overwrite: true
+        displayName: "Copy all files to remote machine"
+        timeoutInMinutes: 10
+      - task: SSH@0
+        inputs:
+          sshEndpoint: $(end_point)
+          runOptions: commands
+          commands: cd "\tmp\nnitest\$(Build.BuildId)" && powershell.exe -command "conda activate l2w | .\uninstall.ps1 | .\install.ps1"
+          failOnStdErr: false
+        displayName: "install on remote windows"
+      - script: python3 -m pip install --upgrade pip setuptools --user
+        displayName: "Install python tools"
+      - script: make easy-install
+        displayName: "Install nni via source code"
+      - script: |
+          sudo apt-get install swig -y
+          PATH=$HOME/.local/bin:$PATH nnictl package install --name=SMAC
+          PATH=$HOME/.local/bin:$PATH nnictl package install --name=BOHB
+        displayName: "Install dependencies for integration tests in remote mode"
+      - script: |
+          set -e
+          cd test
+          python3 nni_test/nnitest/generate_ts_config.py --ts remote --remote_user $(remote_user) --remote_host $(remote_host) \
+          --remote_port $(remote_port) --remote_pwd $(remote_pwd) --nni_manager_ip $(nni_manager_ip)
+          cat config/training_service.yml
+          PATH=$HOME/.local/bin:$PATH python3 nni_test/nnitest/run_tests.py --config config/integration_tests.yml --ts remote
+        displayName: "integration test"
+      - task: SSH@0
+        inputs:
+          sshEndpoint: $(end_point)
+          runOptions: commands
+          commands: rmdir /s /q "\\?\c:\tmp\nnitest\$(Build.BuildId)"
+        condition: always()
+        displayName: "clean up on remote server"

tools/nni_cmd/launcher.py

@@ -139,7 +139,9 @@ def set_remote_config(experiment_config, port, config_file_name):
         for i in range(len(request_data['machine_list'])):
             if isinstance(request_data['machine_list'][i].get('gpuIndices'), int):
                 request_data['machine_list'][i]['gpuIndices'] = str(request_data['machine_list'][i].get('gpuIndices'))
-    response = rest_put(cluster_metadata_url(port), json.dumps(request_data), REST_TIME_OUT)
+    # It needs to connect all remote machines, the time out of connection is 30 seconds.
+    # So timeout of this place should be longer.
+    response = rest_put(cluster_metadata_url(port), json.dumps(request_data), 60, True)
     err_message = ''
     if not response or not check_response(response):
         if response is not None:

tools/nni_cmd/nnictl_utils.py

@@ -227,7 +227,7 @@ def stop_experiment(args):
         experiment_config = Experiments()
         experiment_dict = experiment_config.get_all_experiments()
         for experiment_id in experiment_id_list:
-            print_normal('Stoping experiment %s' % experiment_id)
+            print_normal('Stopping experiment %s' % experiment_id)
             nni_config = Config(experiment_dict[experiment_id]['fileName'])
             rest_pid = nni_config.get_config('restServerPid')
             if rest_pid:

tools/nni_trial_tool/constants.py

@@ -7,8 +7,6 @@ API_ROOT_URL = '/api/v1/nni-pai'
 
 BASE_URL = 'http://{}'
 
-HOME_DIR = os.path.join(os.environ['HOME'], 'nni')
-
 LOG_DIR = os.environ['NNI_OUTPUT_DIR']
 
 NNI_PLATFORM = os.environ['NNI_PLATFORM']

tools/nni_trial_tool/trial_keeper.py

@@ -2,23 +2,27 @@
 # Licensed under the MIT license.
 
 import argparse
-import os
-from subprocess import Popen
-import time
+import ctypes
+import json
 import logging
-import shlex
+import os
 import re
+import shlex
 import sys
-import json
 import threading
-from pyhdfs import HdfsClient
+import time
+from subprocess import Popen
+
 import pkg_resources
-from .rest_utils import rest_post, rest_get
-from .url_utils import gen_send_version_url, gen_parameter_meta_url
+from pyhdfs import HdfsClient
 
-from .constants import LOG_DIR, NNI_PLATFORM, MULTI_PHASE, NNI_TRIAL_JOB_ID, NNI_SYS_DIR, NNI_EXP_ID
-from .hdfsClientUtility import copyDirectoryToHdfs, copyHdfsDirectoryToLocal, copyHdfsFileToLocal
-from .log_utils import LogType, nni_log, RemoteLogger, StdOutputType
+from .constants import (LOG_DIR, MULTI_PHASE, NNI_EXP_ID, NNI_PLATFORM,
+                        NNI_SYS_DIR, NNI_TRIAL_JOB_ID)
+from .hdfsClientUtility import (copyDirectoryToHdfs, copyHdfsDirectoryToLocal,
+                                copyHdfsFileToLocal)
+from .log_utils import LogType, RemoteLogger, StdOutputType, nni_log
+from .rest_utils import rest_get, rest_post
+from .url_utils import gen_parameter_meta_url, gen_send_version_url
 
 logger = logging.getLogger('trial_keeper')
 regular = re.compile('v?(?P<version>[0-9](\.[0-9]){0,1}).*')
@@ -80,6 +84,10 @@ def main_loop(args):
     if hdfs_client is not None:
         copyHdfsDirectoryToLocal(args.nni_hdfs_exp_dir, os.getcwd(), hdfs_client)
 
+    if args.job_id_file:
+        with open(args.job_id_file, 'w') as job_file:
+            job_file.write("%d" % os.getpid())
+
     # Notice: We don't appoint env, which means subprocess wil inherit current environment and that is expected behavior
     log_pipe_stdout = trial_syslogger_stdout.get_pipelog_reader()
     process = Popen(args.trial_command, shell=True, stdout=log_pipe_stdout, stderr=log_pipe_stdout)
@@ -91,6 +99,9 @@ def main_loop(args):
         retCode = process.poll()
         # child worker process exits and all stdout data is read
         if retCode is not None and log_pipe_stdout.set_process_exit() and log_pipe_stdout.is_read_completed == True:
+            # In Windows, the retCode -1 is 4294967295. It's larger than c_long, and raise OverflowError.
+            # So covert it to int32.
+            retCode = ctypes.c_long(retCode).value
             nni_log(LogType.Info, 'subprocess terminated. Exit code is {}. Quit'.format(retCode))
             if hdfs_output_dir is not None:
                 # Copy local directory to hdfs for OpenPAI
@@ -218,6 +229,7 @@ if __name__ == '__main__':
     PARSER.add_argument('--webhdfs_path', type=str, help='the webhdfs path used in webhdfs URL')
     PARSER.add_argument('--nni_manager_version', type=str, help='the nni version transmitted from nniManager')
     PARSER.add_argument('--log_collection', type=str, help='set the way to collect log in trialkeeper')
+    PARSER.add_argument('--job_id_file', type=str, help='set job id file for operating and monitoring job.')
     args, unknown = PARSER.parse_known_args()
     if args.trial_command is None:
         exit(1)